—— Why SignalWire ——

Everyone else stitched two stacks together. We built one.

SignalWire is the only control plane where Voice AI and telephony run on the same real-time media engine; built by the team behind FreeSWITCH.

Talk to us

Read the architecture

The category argument · PUC

CPaaS, UCaaS, and CCaaS are each half a platform

Programmable Unified Communications is what you get when one company owns the call, the channels, and the programming model. Nobody else built it because nobody else owned the media engine to build it on.

CPaaS

INCOMPLETE

APIs without the interaction

Stateless message-passing. You compose primitives, but the call itself is somebody else's problem.

UCaaS

INCOMPLETE

A finished product you can't reshape

Configurable, not programmable. The seams are closed; the experience is the vendor's, not yours.

CCaaS

INCOMPLETE

CC logic bolted to a phone tree

Agents, queues, and IVRs but no first-class programmable surface for the call, the channel, or the AI.

WHAT WE BUILT

LAYER 01

PUC: Programmable Unified Communications

One control plane for voice, video, messaging, SIP, and AI. State lives inside the call. You write the logic; we own the interaction underneath.

STATEFUL

Calls are first-class objects, not event streams.

UNIFIED

Voice + video + messaging + AI in one model.

ELASTIC

Scale across 30+ AZs without rewiring routes.

OWNED

We built it. We run it. From the SBC up.

The category argument · Voice AI

The AI middleware layer is a temporary stack

The orchestration vendors sit on top of carriers they don't own and models they don't host. They charge a tax for the privilege of stitching the media engine together. That layer disappears when somebody finally embeds AI inside the media engine.

The orchestration stack

FOUR LAYERS, FOUR VENDORS

Your application

You

AI orchestration

Vapi · Retell

Real-time rooms

LiveKit

Carrier / SIP / PSTN

Twilio · Telnyx · Bandwidth

Result: latency you can't tune. Governance scattered across four vendors. A surcharge on every minute. A stack with no owner when it breaks.

Platform

ONE VENDOR · ONE STACK

Your application

Control plane: SWML, Agents, Datasphere, PGI

AI kernel: embedded in the media path

Media engine: SIP / PSTN / WebRTC / media

Result: architecture-driven latency. Governance as a platform property. No surcharge on minutes the orchestrator didn't generate. One vendor, one answer. We wrote the code, so we fix it.

The orchestration tax

Voice AI middleware adds a margin on every minute, then routes it through a media engine they don't own. It's a temporary stack: the moment AI is embedded in the media engine, the middleman has nothing left to broker.

The architecture

Media engine. AI kernel. Control plane.

Owning all three is the whole argument. Latency becomes architectural. AI becomes embedded. Governance becomes structural. You can't bolt this on afterwards as it has to be built from the SBC up.

LAYER 03

Control plane

SWML · Agents SDK · Relay · Datasphere · PGI

Declarative call flow, conversation, RAG, and step-scoped tool governance — all one programming model.

LAYER 03

Control plane

SWML · Agents SDK · Relay · Datasphere · PGI

Declarative call flow, conversation, RAG, and step-scoped tool governance — all one programming model.

THE THING NOBODY ELSE HAS

AI kernel

STT · LLM · TTS · Speech-to-speech, in-process

Embedded inside the media path. The agent and the audio live on the same machine, not across a network hop.

THE THING NOBODY ELSE HAS

AI kernel

STT · LLM · TTS · Speech-to-speech, in-process

Embedded inside the media path. The agent and the audio live on the same machine, not across a network hop.

LAYER 01

Real-time media engine

FreeSWITCH-derived · SIP · WebRTC · PSTN · media core

Twenty years of carrier-grade voice and video. The same lineage powers Apple, Amazon, Zoom, and millions of servers.

LAYER 01

Real-time media engine

FreeSWITCH-derived · SIP · WebRTC · PSTN · media core

Twenty years of carrier-grade voice and video. The same lineage powers Apple, Amazon, Zoom, and millions of servers.

Latency is architectural

No network hop between the LLM and the audio. ~1,200ms typical, not 2.5–4s.

AI is embedded

STT, LLM, TTS, and speech-to-speech all live in the kernel, not bolted to the side.

Governance is structural

Tool calls scoped to call state, not prompt-engineered. PGI enforces it at every step.

What this lets you do

Capabilities competitors can't match

When you own the media engine, the AI kernel, and the control plane, certain capabilities aren't features, they're consequences. These are the four that matter.

Stateful interactions across every channel

SIP, WebRTC, PSTN as one programming model. State follows the call, not the channel. You write the logic once; the platform makes it work everywhere.

Voice AI with architectural latency

The kernel and the audio share a process. No queue, no hop, no orchestration tax. ~1,200ms typical end-to-end. It's measurable, reproducible, open-source-verifiable.

AI governance as a platform property

Prompt-engineering is not engineering. PGI (Programmable Governance Interface) scopes which tools an agent can call at each step. The runtime enforces it, not the prompt.

Voice, video, messaging, and AI through one control plane

One contract. One bill. One vendor, one answer. We wrote the code, so we fix it. The control plane is what makes the channels feel like one product instead of four integrations.

20+

Years of innovation

2,700+

Companies on the platform

2.7B

Minutes + messages annually

99.95%

Platform uptime

How we think

Tenets, not tagline

The page is supposed to give you our POV. Here it is, unvarnished. If any of these don't sound right to you, we probably aren't the right vendor.

Prompt and pray is not engineering.

Stochastic glue between deterministic systems is a hack, not a platform. Governance has to live in the runtime, not the prompt.

AI governance is architectural, not a feature.

You can't add it later by editing a system prompt. It's either a property of the platform or it's a marketing claim.

Communications needs a control plane.

Routes, sessions, state, channels, and AI all need to be programmable through one contract or they'll be programmable through none.

Owning the media engine is what makes AI feel real.

The reason conversational AI feels slow is the network hops nobody talks about. Embed the kernel; the latency disappears.

The orchestration layer is temporary.

Middleware that brokers other vendors' media engines lasts exactly until one of those vendors absorbs it. We absorbed it ourselves.

Customer proof

The teams running real traffic through this

These are the teams running real production traffic through SignalWire, from healthcare platforms to AI-native startups to contact-center operators.

“There’s no magic about what Twilio was doing - spending two hours switching to SignalWire was a no brainer.”
Mauricio Sastre
VP of Product
“Connecting to different carriers is complicated, but SignalWire’s team of industry experts makes BYOC easy.”
Mauricio Sastre
VP of Product
“Switching to SignalWire was really fast. It took us maybe a day to get a proof of concept working. We were able to get up and running really quickly...The level of knowledge and support that SignalWire offers is leaps and bounds ahead of the competition.”
Mark Lilien
Co-founder and CTO

“There’s no magic about what Twilio was doing - spending two hours switching to SignalWire was a no brainer.”
Mauricio Sastre
VP of Product
“Connecting to different carriers is complicated, but SignalWire’s team of industry experts makes BYOC easy.”
Mauricio Sastre
VP of Product
“Switching to SignalWire was really fast. It took us maybe a day to get a proof of concept working. We were able to get up and running really quickly...The level of knowledge and support that SignalWire offers is leaps and bounds ahead of the competition.”
Mark Lilien
Co-founder and CTO

Heritage

Built by the team that built FreeSWITCH

Twenty years of real-time voice infrastructure. The same engine powers Apple, Amazon, Zoom, and millions of servers worldwide. The people who wrote it built the media engine underneath SignalWire.

Telecom insiders, refined for cloud-native developers. That's not marketing; it's how latency works.

About the team

FreeSWITCH today

FROM FREESWITCH TO THE CONTROL PLANE

2005

Anthony Minessale starts FreeSWITCH

The open-source telephony engine that becomes the media engine.

2017

SignalWire founded by the core team

Cloud-native, programmable, the same engine modernized.

2020

PUC platform — voice, video, messaging, SIP

One control plane across every channel.

2023

AI kernel — STT + LLM + TTS in-process

The media engine gets a brain.

2026

Speech-to-speech, same kernel

No pipeline. Just the model and the audio.

FAQS

Frequently asked questions

Everything you might want to know about what makes SignalWire different.

What is SignalWire?

SignalWire is a programmable communications platform built on a real-time media engine we own: voice, video, messaging, SIP, and Voice AI through one control plane. We were founded by the FreeSWITCH team and run the same engineering lineage end-to-end.

What is PUC, and why does it matter?

PUC (Programmable Unified Communications) is the category we built. CPaaS gives you APIs without owning the interaction. UCaaS gives you a finished product you can't reshape. PUC gives you both: a programmable surface that owns the call, the channels, and the AI underneath them.

What makes SignalWire a Voice AI platform?

The AI kernel is embedded in the media path, not bolted on top. STT, LLM, TTS, and speech-to-speech all run inside the media engine. That's why typical conversational latency is around 1,200ms and why governance is a property of the runtime instead of a system prompt.

How is SignalWire different from Twilio?

Twilio exposes APIs over a media stack they don't fully own. We own the media engine, the AI kernel, and the control plane. It gives you one stack, one contract, one programming model across every channel. For a side-by-side, see signalwire.com/compare/twilio-alternative.

How is SignalWire different from Vapi, Retell, or LiveKit?

Those vendors orchestrate other companies' stacks (Vapi, Retell) or own great rooms-and-video infrastructure without the carrier-grade phone media engine (LiveKit). We own the entire media engine they orchestrate or borrow from. The orchestration layer is a temporary stack; we removed it ourselves.

Who built SignalWire?

The team that built FreeSWITCH, the open-source real-time voice engine that powers Apple, Amazon, Zoom, and millions of servers worldwide. Founded in 2017 by Anthony Minessale and the core FreeSWITCH engineers.

How does SignalWire's AI governance work?

Governance is enforced by the runtime, not the prompt. PGI (Programmable Governance Interface) scopes which tools an agent can call at each step. The runtime enforces it, not the prompt.The audit trail is structural. Prompt-and-pray is not engineering and we don't treat it as such.

Is SignalWire a CPaaS?

No. CPaaS describes platforms that expose programmable APIs on top of someone else's media engine. We built our own media engine and embedded AI inside it. PUC is the category that captures what we actually are.

Ready to ship?

Sign up free and have a voice agent on the line by the end of the day. SWML, the Agents SDK, and the Skills library are all open to you from the first request.

Start building

Dive into docs

Ready to ship?

Sign up free and have a voice agent on the line by the end of the day. SWML, the Agents SDK, and the Skills library are all open to you from the first request.

Start building

Dive into docs

Ready to ship?

Sign up free and have a voice agent on the line by the end of the day. SWML, the Agents SDK, and the Skills library are all open to you from the first request.

Start building

Dive into docs

Ready to ship?

Sign up free and have a voice agent on the line by the end of the day. SWML, the Agents SDK, and the Skills library are all open to you from the first request.

Start building

Dive into docs